A safety risk early warning method and system for sports training

Abstract

The application provides a safety risk early warning method and system for sports training, which comprises the following steps: performing real-time body movement analysis on all monitored objects based on a sports training monitoring video to obtain body movement analysis results; predicting the body movement trajectory of the monitored objects based on the body movement analysis results to obtain body movement dynamic prediction results; evaluating the current safety risk coefficient of the monitored objects based on the body movement dynamic prediction results; judging whether a safety early warning instruction needs to be sent based on the current safety risk coefficient to obtain a safety risk early warning result; and analyzing the body movement of the monitored objects based on the sports training monitored objects, predicting the body movement trajectory of the monitored objects based on the body movement analysis results, evaluating the safety risk coefficient based on the prediction results, and then judging whether to perform safety early warning, so that timely, accurate and effective safety risk early warning in the sports training process is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk early warning, in particular to a safety risk early warning method and system for sports training. BACKGROUND

[0002] At present, the safety guarantee of sports training is mainly achieved by manually supervising the possible dangerous actions in real time on site to realize the risk early warning of the actions of personnel who are not standardized or have safety risks, or by manually inspecting sports training equipment to realize the safety early warning of safety risk accidents caused by the failure of sports training equipment.

[0003] However, whether it is manual supervision or manual inspection, in addition to the defect of the need for manual cost investment, there are also defects of time lag of early warning and error of safety risk judgment.

[0004] Therefore, the present application provides a safety risk early warning method and system for sports training. SUMMARY

[0005] The present application provides a safety risk early warning method and system for sports training, which analyzes the body movements of monitoring objects based on sports training monitoring objects, predicts the body movement trajectories of the monitoring objects based on the body movement analysis results, evaluates the safety risk coefficients based on the prediction results, and then judges whether to perform safety early warning, so as to overcome the defects of time lag and judgment error of the traditional safety risk early warning method of sports training without the need for manual cost investment, and to realize timely, accurate and effective safety risk early warning in the process of sports training.

[0006] The present application provides a safety risk early warning method for sports training, comprising:

[0007] S1: performing real-time body movement analysis on all monitoring objects based on sports training monitoring videos to obtain body movement analysis results;

[0008] S2: predicting the body movement trajectories of the monitoring objects based on the body movement analysis results to obtain body movement dynamic prediction results;

[0009] S3: evaluating the current safety risk coefficients of the monitoring objects based on the body movement dynamic prediction results;

[0010] S4: judging whether to issue a safety early warning instruction based on the current safety risk coefficients to obtain a safety risk early warning result.

[0011] Preferably, the safety risk early warning method for sports training, S1: performing real-time body movement analysis on all monitoring objects based on sports training monitoring videos to obtain body movement analysis results, comprises:

[0012] S101: acquiring a sports training monitoring video of a current sports training process in real time;

[0013] S102: determining a difference region between a remaining video frame and a corresponding adjacent previous video frame in the sports training monitoring video except for a first video frame;

[0014] S103: screening a plurality of difference regions determined to belong to a same object from all the difference regions in the sports training monitoring video, and regarding a sequence of the plurality of difference regions continuously determined to belong to the same object as a same object region sequence;

[0015] S104: calculating a similarity between each same object region sequence and each limb posture graph in a limb image library, regarding a same object region sequence with a similarity not less than a similarity threshold as a limb motion region sequence, and regarding all the limb motion region sequences as a limb motion analysis result.

[0016] Preferably, the safety risk early warning method for sports training screens a plurality of difference regions determined to belong to a same object from all the difference regions in the sports training monitoring video, which comprises:

[0017] detecting all contour corner points in a contour of the difference region based on a corner detection algorithm, determining adjacent contour points of the contour corner points, and determining a tangent angle of the contour of the difference region at the adjacent contour points;

[0018] determining a contour line length between two adjacent contour points of the contour corner points, regarding a difference value between the tangent angles at the two adjacent contour points of the contour corner points and a ratio of the corresponding contour line length as an approximate arc curvature of the contour corner points;

[0019] regarding a line connecting the center point of the difference region and an adjacent contour point of the contour corner points in a counterclockwise direction of the contour of the difference region as a first line, and regarding a line connecting the center point of the difference region and an adjacent contour point of the contour corner points in a clockwise direction of the contour of the difference region as a second line;

[0020] regarding an included angle between the first line and the second line corresponding to the contour corner points as an approximate arc included angle of the contour corner points;

[0021] screening the plurality of difference regions determined to belong to the same object from all the difference regions in the sports training monitoring video based on the approximate arc curvature and the approximate arc included angle of all the contour corner points in all the difference regions.

[0022] Preferably, the safety risk early warning method for sports training comprises the following steps:

[0023] sequencing the approximate arc curvatures of the contour corner points in the difference region in a counterclockwise direction to obtain an approximate arc curvature sequence, and sequencing the approximate arc angles of the contour corner points in the difference region in the counterclockwise direction to obtain an approximate arc angle sequence;

[0024] calculating the similarity between the two difference regions in the two adjacent video frames based on the average chroma of the difference region and the approximate arc curvature sequence and the approximate arc angle sequence;

[0025] judging the two difference regions with a similarity not less than a similarity threshold as belonging to the same object to obtain a same-judgment result;

[0026] screening the multiple difference regions judged as belonging to the same object from all the difference regions in the sports training monitoring video based on the same-judgment result.

[0027] Preferably, the safety risk early warning method for sports training comprises the following steps S2: predicting the limb movement trajectory of the monitoring object based on the limb movement analysis result to obtain a limb movement dynamic prediction result, which comprises the following steps:

[0028] S201: performing contour recognition on the video frames in the sports training monitoring video to obtain all the contours in each video frame;

[0029] S202: determining the monitoring object corresponding to each limb movement region sequence in the limb movement analysis result based on all the contours in the video frames and the contours of all the difference regions in the limb movement region sequence;

[0030] S203: determining the limb name of each limb movement region sequence based on the distribution positions of all the difference regions belonging to the same video frame in all the limb movement region sequences of the same monitoring object;

[0031] S204: determining the limb movement trajectory corresponding to each limb movement region sequence based on the coordinate representation of each difference region in the limb movement region sequence;

[0032] S205: predicting the limb movement trajectory of the monitoring object based on the limb name and the limb movement trajectory of all the limb movement region sequences of the monitoring object and a preset trajectory prediction model to obtain a limb movement dynamic prediction result.

[0033] Preferably, the safety risk early warning method for sports training, S3: based on the limb action dynamic prediction result, the current safety risk coefficient of the monitored object is evaluated, comprising:

[0034] It is judged whether there is action interaction between all monitored objects in the sports training monitoring video. If yes, a multi-party action dynamic prediction interaction model of all monitored objects is built;

[0035] Based on the multi-party action dynamic prediction interaction model, the current safety risk coefficient of the monitored object with action interaction is analyzed, and based on the limb action dynamic prediction result of the monitored object without action interaction, the current safety risk coefficient of the monitored object without action interaction is analyzed;

[0036] Otherwise, the current safety risk coefficient of the monitored object is analyzed based on the limb action dynamic prediction result.

[0037] Preferably, the safety risk early warning method for sports training, based on the multi-party action dynamic prediction interaction model, the current safety risk coefficient of the monitored object with action interaction is analyzed, comprising:

[0038] Based on the multi-party action dynamic prediction interaction model, the action initiator and the action receiver of each action are determined;

[0039] Based on all the limb action dynamic prediction results and the action initiator and the action receiver of all actions, the current safety risk coefficient of all monitored objects is analyzed.

[0040] Preferably, the safety risk early warning method for sports training, S4: based on the current safety risk coefficient, it is judged whether the safety early warning instruction needs to be issued, and the safety risk early warning result is obtained, comprising:

[0041] When there is a current safety risk coefficient exceeding the safety risk coefficient threshold, the safety early warning instruction is issued;

[0042] When there is no current safety risk coefficient exceeding the safety risk coefficient threshold, the judgment result is retained.

[0043] Preferably, the safety risk early warning method for sports training further comprises:

[0044] S5: The source, inspection record and use record of the sports training equipment are registered, and the safety account book of the sports training equipment is generated;

[0045] S6: Based on the safety account book of the sports training equipment, the equipment safety coefficient of the sports training equipment is evaluated, based on the equipment safety coefficient and the equipment safety coefficient threshold, it is judged whether the safety early warning instruction needs to be issued, and the equipment safety risk early warning result is obtained.

[0046] The application provides a safety risk early warning system for sports training, comprising:

[0047] An action analysis module is configured to perform real-time body action analysis on all monitored objects based on the sports training monitoring video, and obtain body action analysis results;

[0048] A trajectory prediction module is configured to predict the body action trajectory of the monitored object based on the body action analysis results, and obtain body action dynamic prediction results;

[0049] A risk assessment module is configured to assess the current safety risk coefficient of the monitored object based on the body action dynamic prediction results;

[0050] A safety warning module is configured to determine whether a safety warning instruction needs to be sent based on the current safety risk coefficient, and obtain a safety risk early warning result.

[0051] Other features and advantages of the present application will be described in the following description, and some will become apparent from the description, or will be learned through practice of the application. The purpose and other advantages of the present application can be achieved and obtained by the structure specifically pointed out in the written description, claims, and drawings.

[0052] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation on the present application. In the drawings:

[0054] Figure 1 A flow chart of a safety risk early warning method for sports training in an embodiment of the present application;

[0055] Figure 2 A flow chart of another safety risk early warning method for sports training in an embodiment of the present application;

[0056] Figure 3 A flow chart of still another safety risk early warning method for sports training in an embodiment of the present application;

[0057] Figure 4 A schematic diagram of a safety risk early warning system for sports training in an embodiment of the present application. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present application will be described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application.

[0059] Embodiment 1:

[0060] The application provides a safety risk early warning method for sports training, which refers to Figure 1 , comprising:

[0061] S1: performing real-time body movement analysis on all monitoring objects based on a sports training monitoring video to obtain a body movement analysis result;

[0062] S2: predicting a body movement trajectory of the monitoring object based on the body movement analysis result to obtain a body movement dynamic prediction result;

[0063] S3: evaluating a current safety risk coefficient of the monitoring object based on the body movement dynamic prediction result;

[0064] S4: determining whether a safety early warning instruction needs to be issued based on the current safety risk coefficient to obtain a safety risk early warning result.

[0065] In this embodiment, the sports training monitoring video is a monitoring video of a sports training process of the monitoring object.

[0066] In this embodiment, the monitoring object is an object monitored based on the safety risk early warning method for sports training provided in this embodiment.

[0067] In this embodiment, real-time body movement analysis is a process of analyzing the movement of the monitoring object in real time.

[0068] In this embodiment, the body movement analysis result is a result obtained by performing real-time body movement analysis on the monitoring object, for example, including a change sequence (represented by a series of body regions) of a body region corresponding to a movement process of a body of the monitoring object in the sports training monitoring video.

[0069] In this embodiment, the body movement trajectory is a movement trajectory of a body in a single movement process.

[0070] In this embodiment, the body movement dynamic prediction result is a result obtained by predicting the body movement trajectory of the monitoring object based on the body movement analysis result, that is, a movement trajectory of a movement possibly made by the body of the monitoring object in the next time, which can be a continuation of the current movement or a new movement.

[0071] In this embodiment, the current safety risk coefficient is a current safety risk coefficient of the monitoring object evaluated based on the body movement dynamic prediction result, and the safety risk coefficient is a numerical value representing the danger degree of the movement made by the monitoring object.

[0072] In this embodiment, the safety warning instruction is an instruction for reminding the monitored object or a user responsible for the safety risk supervision task of the sports training process that a safety risk problem will occur in the current sports training process.

[0073] In this embodiment, the safety risk warning result is a result of determining whether to issue a safety warning instruction based on the current safety risk coefficient.

[0074] The above technology has the beneficial effect that the body movement of the monitored object is analyzed based on the sports training monitoring object, the body movement trajectory of the monitored object is predicted based on the body movement analysis result, the safety risk coefficient is evaluated based on the prediction result, and it is determined whether to perform safety warning, without the need for human cost investment, overcoming the time lag and judgment error defects of the traditional safety risk warning method of sports training, and realizing timely, accurate and effective safety risk warning in the sports training process.

[0075] Embodiment 2:

[0076] Based on embodiment 1, the safety risk warning method for sports training, S1: performing real-time body movement analysis on all monitored objects based on sports training monitoring video, obtaining body movement analysis result, referring to Figure 2 , comprising:

[0077] S101: Real-time acquisition of sports training monitoring video of the current sports training process;

[0078] S102: Determining the difference area of the remaining video frames in the sports training monitoring video except the first video frame and the corresponding adjacent previous video frame;

[0079] S103: Selecting a plurality of difference areas determined to belong to the same object from all difference areas in the sports training monitoring video, and regarding the sequence of a plurality of difference areas continuously determined to belong to the same object as a same object area sequence;

[0080] S104: Calculating the similarity of each same object area sequence and each body posture graph in the body image library, and regarding the same object area sequence with a similarity not less than a similarity threshold as a body movement area sequence, and regarding all body movement area sequences as body movement analysis results.

[0081] In this embodiment, the difference area is the area in the remaining video frames in the sports training monitoring video except the first video frame that is different from the corresponding adjacent previous video frame (i.e., the video frame adjacent to the two video frames, the video frame located in the previous video frame of the current video frame).

[0082] In this embodiment, the same object is the same physical object (the object can be a movable object or a human body limb part).

[0083] In this embodiment, the same object region sequence is a sequence of the difference regions continuously determined as belonging to the same object, i.e., for example, when the difference region a1 in the second video frame and the difference region a3 in the third video frame are determined as belonging to the same object, and the difference region a3 in the third video frame and the difference region a4 in the fourth video frame are determined as belonging to the same object, then the sequence of the difference region a1, the difference region a2, and the difference region a3 in order is taken as the same object region sequence.

[0084] In this embodiment, the same object region sequence is a sequence of the difference regions continuously determined as belonging to the same object.

[0085] In this embodiment, the limb image library is a library for storing a plurality of posture images of each limb part of a human body.

[0086] In this embodiment, the limb posture image is an image containing a certain posture of a certain limb of a human body.

[0087] In this embodiment, the similarity between each same object region sequence and each limb posture image in the limb image library is calculated, including:

[0088] The similarity between each same object region sequence and each limb posture image in the limb image library is calculated based on the tangent angle of the contour of the difference region in each same object region sequence at each point in the corresponding contour and the tangent angle of the contour of each limb posture image in the limb image library at each point in the corresponding contour.

[0089]

[0090] In the formula, s1 is the similarity between the currently calculated difference region in the currently calculated same object region sequence and the currently calculated limb posture image in the limb image library, n is the total number of points in the contour of the currently calculated difference region, m is the total number of points in the contour of the currently calculated limb posture image, α i is the tangent angle of the contour of the currently calculated difference region in the currently calculated same object region sequence at the i-th point in the corresponding contour, and α j is the tangent angle of the contour of the currently calculated limb posture image at the j-th point in the corresponding contour.

[0091] The average of the similarities between all the difference regions in the currently calculated same object region sequence and the currently calculated limb posture image in the limb image library is taken as the similarity between the currently calculated same object region sequence and the currently calculated limb posture image in the limb image library.

[0092] Based on the above process, the similarity between each same object region sequence and each limb posture graph in the limb image library can be accurately calculated based on the tangent angle of each point in the corresponding contour of the contour of the difference region in each same object region sequence and the tangent angle of each point in the corresponding contour of the contour of each limb posture graph in the limb image library.

[0093] In this embodiment, the similarity threshold is the threshold of the similarity used for screening the limb action region sequence in the same object region sequence.

[0094] In this embodiment, the limb action region sequence is a sequence of limb regions that belong to the action process of the same limb part.

[0095] The above technology has the beneficial effect that by determining the same object region sequence that belongs to the same real object in the difference region of the adjacent video frames of the sports training monitoring video, the region sequence that represents the movement process of the moving real object is preliminarily screened, and by comparing the similarity between the same object region sequence and each limb posture graph in the limb image library with the similarity threshold, the limb action region sequence is screened, thereby realizing the screening of the change region sequence corresponding to the limb part that makes the action in the moving real object.

[0096] Embodiment 3:

[0097] On the basis of embodiment 2, the safety risk early warning method for sports training screens a plurality of difference regions that are determined to belong to the same object from all the difference regions in the sports training monitoring video, and refers to the similarity threshold of the similarity between each same object region sequence and each limb posture graph in the limb image library, and includes: Figure 3 , comprising:

[0098] Based on the corner point detection algorithm, all contour corner points of the contour of the difference region are detected, the adjacent contour points of the contour corner points are determined, and the tangent angle of the contour of the difference region at the adjacent contour points is determined;

[0099] The contour line length between the two adjacent contour points of the contour corner points is determined, and the difference value of the tangent angles at the two adjacent contour points of the contour corner points is taken as the approximate curvature of the contour corner points.

[0100] The line connecting the center point of the difference region and the adjacent contour point of the contour corner point in the counterclockwise direction of the contour of the difference region is taken as the first connecting line, and the line connecting the center point of the difference region and the adjacent contour point of the contour corner point in the clockwise direction of the contour of the difference region is taken as the second connecting line;

[0101] The included angle between the first connecting line and the second connecting line corresponding to the contour corner point is taken as the approximate arc included angle of the contour corner point.

[0102] Based on the approximate arc curvature and the approximate arc included angle of all contour corner points in all difference regions in the sports training monitoring video, multiple difference regions determined to belong to the same object are screened out in all difference regions.

[0103] In this embodiment, the corner point detection algorithm is, for example, a Harris corner point detection algorithm.

[0104] In this embodiment, the contour corner point is a corner point detected in the contour of the difference region based on the corner point detection algorithm.

[0105] In this embodiment, the adjacent contour point is a contour point adjacent to the contour corner point in the contour of the difference region.

[0106] In this embodiment, the approximate arc curvature is the approximate curvature of the part of the contour of the difference region at a certain contour corner point, which is approximately an arc or a curve segment in the shape of an arc (the curve segment or the arc is between the adjacent contour points of the contour corner point).

[0107] In this embodiment, the center point is a point position corresponding to the average value of the coordinates of all points in the difference region.

[0108] In this embodiment, the approximate arc included angle is the included angle between the contour of the corresponding difference region at the contour corner point and the line connecting the adjacent contour points of the contour corner point and the center point of the difference region.

[0109] The beneficial effects of the above technology are: based on the contour line length between the adjacent contour points of the contour corner point in the contour of the difference region and the difference value between the tangent angles of the contour at the adjacent contour points, the approximate arc curvature of the part of the contour of the difference region at the contour corner point is calculated in the thinking of curvature (i.e., the approximate calculation of the arc curvature), and based on the angle between the contour of the contour corner point in the difference region and the line connecting the adjacent contour points and the center point of the difference region, the approximate arc included angle is calculated (i.e., the approximate calculation of the included angle between the two ends of the arc and the line connecting the region center point), the shape of the difference region is numerically processed based on the approximate arc curvature and the approximate arc included angle calculated above, and then the shape analysis of the difference region is realized based on the approximate arc curvature and the approximate arc included angle of the contour corner point, and multiple difference regions determined to belong to the same object are screened out in all difference regions.

[0110] Embodiment 4:

[0111] Based on embodiment 3, the safety risk early warning method for sports training, based on the approximate arc curvature and the approximate arc included angle of all contour corner points in all difference regions in the sports training monitoring video, multiple difference regions determined to belong to the same object are screened out in all difference regions, comprising:

[0112] sequentially arranging the approximate arc curvatures of all the contour corner points in the difference region in the counterclockwise direction to obtain an approximate arc curvature sequence, and sequentially arranging the approximate arc included angles of all the contour corner points in the difference region in the counterclockwise direction to obtain an approximate arc included angle sequence;

[0113] calculating the similarity between the two difference regions belonging to two adjacent video frames based on the average chroma of the difference region and the approximate arc curvature sequence and the approximate arc included angle sequence;

[0114] judging the two difference regions with the similarity not less than the similarity threshold as belonging to the same object to obtain a same-subject judgment result;

[0115] screening, based on the same-subject judgment result, the multiple difference regions judged as belonging to the same object from all the difference regions in the sports training monitoring video.

[0116] In this embodiment, the approximate arc curvature sequence is the sequence obtained by sequentially arranging the approximate arc curvatures of all the contour corner points in the counterclockwise direction, that is, the order of the contour corner points corresponding to the order of the approximate arc curvatures in the approximate arc curvature sequence is the order of the contour corner points sequentially distributed in the counterclockwise direction in the contour of the difference region.

[0117] In this embodiment, the approximate arc included angle sequence is the sequence obtained by sequentially arranging the approximate arc included angles of all the contour corner points in the counterclockwise direction, that is, the order of the contour corner points corresponding to the order of the approximate arc included angles in the approximate arc included angle sequence is the order of the contour corner points sequentially distributed in the counterclockwise direction in the contour of the difference region.

[0118] In this embodiment, the average chroma is the average value of the chroma values of all the pixel points in the difference region.

[0119] In this embodiment, the similarity between the two difference regions belonging to two adjacent video frames is calculated based on the average chroma of the difference region and the approximate arc curvature sequence and the approximate arc included angle sequence, including:

[0120]

[0121] In the formula, s2 is the similarity currently calculated between the two difference regions belonging to two adjacent video frames, A is the average chroma currently calculated of the first difference region among the two difference regions belonging to two adjacent video frames, B is the average chroma currently calculated of the second difference region among the two difference regions belonging to two adjacent video frames, b is the smaller value of the total number of contour corner points of each of the two difference regions currently calculated belonging to two adjacent video frames, max is the larger value of the total number of contour corner points of each of the two difference regions currently calculated belonging to two adjacent video frames, and κ is a constant.a1 is the athapproximate arc curvature in the sequence of approximate arc curvatures belonging to the first of the two difference regions in the two adjacent video frames, calculated currently, κ a2 is the athapproximate arc curvature in the sequence of approximate arc curvatures belonging to the second of the two difference regions in the two adjacent video frames, calculated currently, θ a1 is the athapproximate arc angle in the sequence of approximate arc angles belonging to the first of the two difference regions in the two adjacent video frames, calculated currently, θ a2 is the athapproximate arc angle in the sequence of approximate arc angles belonging to the second of the two difference regions in the two adjacent video frames, calculated currently;

[0122] Based on the above formula, the similarity between the two difference regions in the two adjacent video frames can be accurately calculated based on the average chrominance of the difference region, the sequence of approximate arc curvatures, and the sequence of approximate arc angles.

[0123] In this embodiment, the same-species determination result is the result of determining that the two difference regions with the similarity not less than the similarity threshold belong to the same object.

[0124] In this embodiment, based on the same-species determination result, the multiple difference regions determined to belong to the same object are screened out from all the difference regions in the sports training monitoring video, that is:

[0125] For example, when the difference region a1 in the second video frame and the difference region a3 in the third video frame are determined to belong to the same object in the same-species determination result, and the difference region a3 in the third video frame and the difference region a4 in the fourth video frame are determined to belong to the same object, then the sequence formed by sequentially sorting the difference region a1, the difference region a2, and the difference region a3 is regarded as the same-object region sequence.

[0126] The above technology has the beneficial effects that: based on the sequence of approximate arc curvatures and the sequence of approximate arc angles formed by the approximate arc curvatures and the approximate arc angles of all the contour angle points in the difference region, and the average chrominance of the difference region, the similarity between the two difference regions is accurately calculated from the perspective of the contour shape of the difference region and the comprehensive chrominance value of the region, and then the same-species determination result is obtained based on the calculated similarity, and further, the multiple difference regions belonging to the same object are screened out, that is, the multiple difference regions belonging to the same moving physical object are screened out in the sports training monitoring video, and the matching screening of the regions corresponding to the same moving physical object in different video frames is completed.

[0127] Embodiment 5:

[0128] On the basis of embodiment 1, the safety risk early warning method for sports training, S2: based on the limb action analysis result, the limb action trajectory of the monitored object is predicted, and the limb action dynamic prediction result is obtained, comprising:

[0129] S201: Contour recognition is performed on the video frames in the sports training monitoring video, and all contours in each video frame are obtained;

[0130] S202: Based on all contours in the video frame and contours of all difference regions in the limb action region sequence, the monitored object corresponding to each limb action region sequence in the limb action analysis result is determined;

[0131] S203: Based on the distribution position of all difference regions belonging to the same video frame in all limb action region sequences of the same monitored object, the limb name of each limb action region sequence is determined;

[0132] S204: Based on the coordinate representation of each difference region in the limb action region sequence, the limb movement trajectory corresponding to each limb action region sequence is determined;

[0133] S205: Based on the limb name and limb movement trajectory of all limb action region sequences of the monitored object and the preset trajectory prediction model, the limb action trajectory of the monitored object is predicted, and the limb action dynamic prediction result is obtained.

[0134] In this embodiment, contour recognition is performed on the video frames in the sports training monitoring video based on a contour recognition algorithm (such as the Canny edge detection algorithm).

[0135] In this embodiment, based on all contours in the video frame and contours of all difference regions in the limb action region sequence, the monitored object corresponding to each limb action region sequence in the limb action analysis result is determined, that is:

[0136] Based on the preset human body contour recognition model (i.e. the model for screening out human body contours obtained by training a large number of human body contours in advance), human body contours are identified in all contours;

[0137] The difference region located in the region surrounded by the human body contour in the corresponding video frame is regarded as the limb region of the monitored object corresponding to the corresponding human body contour;

[0138] The total number of difference regions of the limb region belonging to each monitored object (the total number of monitored objects is consistent with the total number of human body contours in a single video frame) in the limb action region sequence is determined, and the monitored object corresponding to the maximum total number is regarded as the monitored object corresponding to the limb action region sequence.

[0139] In this embodiment, the limb name of each limb action region sequence is determined based on the distribution positions (e.g., the highest position, the upper left position, the lower right position, the upper half, the lower half, etc.) of all the difference regions belonging to the same video frame in all the limb action region sequences of the same monitored object, that is:

[0140] The distribution positions of all the difference regions belonging to the same video frame in all the limb action region sequences of the same monitored object and the approximate shapes of all the difference regions are retrieved from the region position shape-limb name list (a preset list containing possible region positions and shapes corresponding to different limb names) to determine the limb name of each difference region.

[0141] In this embodiment, the limb motion trajectory corresponding to each limb action region sequence is determined based on the coordinate representation of each difference region in the limb action region sequence, that is:

[0142] The average value of the coordinate values of all the points in the difference region in the coordinate representation of each difference region in the limb action region sequence is taken as the center point of the corresponding difference region, and the trajectory formed by sequentially sorting the center points of all the difference regions in the limb action region sequence is taken as the corresponding limb motion trajectory.

[0143] In this embodiment, the preset trajectory prediction model is a model for predicting the motion trajectory that may be made next by different limb parts, which is obtained by training a large number of limb motion trajectories of different limb parts in advance.

[0144] In this embodiment, the limb action trajectory of the monitored object is predicted based on the limb names and limb motion trajectories of all the limb action region sequences of the monitored object and the preset trajectory prediction model, and the limb action dynamic prediction result is obtained, that is:

[0145] The limb action prediction trajectory of the corresponding limb name obtained after inputting the limb name and the limb motion trajectory of the limb action region sequence into the preset trajectory prediction model, for example, the input limb name is leg, and the limb motion trajectory is a circular arc, and the corresponding limb action prediction trajectory is a longer and approximately curved circular arc that continues the circular arc of the current limb motion trajectory.

[0146] The limb action prediction dynamic trajectory of all the limb names contained in the monitored object (i.e., a trajectory containing the movement process of the action trajectory in the next time) is taken as the limb action dynamic prediction result.

[0147] The beneficial effects of the above technology are: based on the contours of the video frames in the sports training monitoring video and the contours of the difference regions, the belonging monitoring object of the difference region is accurately determined, and the belonging monitoring object of the limb action region sequence composed of multiple difference regions is further accurately determined, and the limb name is determined combined with the distribution position of the difference region; based on the coordinate representation of the difference region in the limb action region sequence, the limb motion trajectory is determined, and the accurate prediction of the limb motion trajectory is realized combined with the preset trajectory prediction model.

[0148] Embodiment 6:

[0149] Based on the embodiment 1, the safety risk early warning method for sports training, S3: based on the limb action dynamic prediction result, the current safety risk coefficient of the monitoring object is evaluated, including:

[0150] It is judged whether there is action interaction between all monitoring objects in the sports training monitoring video, if yes, a multi-party action dynamic prediction interaction model of all monitoring objects is built;

[0151] Based on the multi-party action dynamic prediction interaction model, the current safety risk coefficient of the monitoring object with action interaction is analyzed, and based on the limb action dynamic prediction result of the monitoring object without action interaction, the current safety risk coefficient of the monitoring object without action interaction is analyzed;

[0152] Otherwise, based on the limb motion action dynamic prediction result, the current safety risk coefficient of the monitoring object is analyzed.

[0153] In this embodiment, the action interaction is that the direct receiver of the action of the monitoring object is another monitoring object.

[0154] In this embodiment, the multi-party action dynamic prediction interaction model is a three-dimensional dynamic model containing the action process currently made by multiple monitoring objects.

[0155] In this embodiment, based on the limb action dynamic prediction result of the monitoring object without action interaction, the current safety risk coefficient of the monitoring object without action interaction is analyzed, or based on the limb motion action dynamic prediction result, the current safety risk coefficient of the monitoring object is analyzed, including:

[0156] The trajectory shape and trajectory average moving speed of the limb action prediction dynamic trajectory in the limb action dynamic prediction result, and the maximum real-time speed of trajectory movement are determined;

[0157] Based on the trajectory shape and trajectory average moving speed of the limb action prediction dynamic trajectory, and the maximum real-time speed of trajectory movement, the current safety risk coefficient of the monitoring object without action interaction is analyzed:

[0158] determine the first safety risk coefficient of the corresponding action based on the safety risk coefficient table of the corresponding limb name according to the trajectory shape;

[0159] take the ratio of the difference between the average moving speed of the trajectory and the preset average moving speed threshold of the corresponding limb name and the corresponding preset average moving speed threshold as the second safety risk coefficient of the corresponding action;

[0160] take the ratio of the difference between the maximum real-time speed and the preset real-time speed threshold of the corresponding limb name and the corresponding preset real-time speed threshold as the third safety risk coefficient of the corresponding action;

[0161] take the maximum value among the first safety risk coefficient, the second safety risk coefficient and the third safety risk coefficient of the action as the safety risk coefficient of the corresponding action;

[0162] take the maximum value among the safety risk coefficients of all actions of the monitoring object as the current safety risk coefficient of the monitoring object.

[0163] The above-mentioned technical beneficial effects are that: based on the judgment result of whether there is action interaction between the monitoring objects, the process of evaluating the current safety risk coefficient of the monitoring object is divided into two cases: when there is action interaction, the current safety risk coefficient of the monitoring object with action interaction is analyzed based on the multi-party action dynamic prediction interaction model; when there is no action interaction, the current safety risk coefficient of the monitoring object is analyzed based on the dynamic prediction result of the limb movement action, which further ensures the accuracy of the evaluated current safety risk coefficient.

[0164] Embodiment 7:

[0165] On the basis of embodiment 6, the safety risk early warning method for sports training analyzes the current safety risk coefficient of the monitoring object with action interaction based on the multi-party action dynamic prediction interaction model, which comprises:

[0166] determining the action initiator and the action receiver of each action based on the multi-party action dynamic prediction interaction model;

[0167] analyzing the current safety risk coefficients of all monitoring objects based on the dynamic prediction result of all limb actions and the action initiator and the action receiver of all actions.

[0168] In this embodiment, the action initiator is the monitoring object that sends out the action based on the multi-party action dynamic prediction interaction model.

[0169] In this embodiment, the action receiver is the monitoring object whose limb movement trajectory intersects with the human body contour based on the multi-party action dynamic prediction interaction model.

[0170] In this embodiment, based on the dynamic prediction result of all limb actions and the action initiator and action receiver of all actions, the current safety risk coefficient of all monitored objects is analyzed, including:

[0171] The trajectory shape and trajectory average moving speed of the limb action prediction dynamic trajectory in the limb action dynamic prediction result, and the maximum real-time speed of trajectory movement are determined.

[0172] Based on the trajectory shape, trajectory average moving speed, and maximum real-time speed of trajectory movement of the limb action prediction dynamic trajectory, the current safety risk coefficient of the monitored object without action interaction is analyzed:

[0173] Based on the trajectory shape, the safety risk coefficient table corresponding to the limb name is searched, and the first safety risk coefficient corresponding to the action is determined.

[0174] The difference between the trajectory average moving speed and the corresponding preset average moving speed threshold value and the ratio of the corresponding preset average moving speed threshold value are taken as the second safety risk coefficient of the corresponding action.

[0175] The difference between the maximum real-time speed and the corresponding preset real-time speed threshold value and the ratio of the corresponding preset real-time speed threshold value are taken as the third safety risk coefficient of the corresponding action.

[0176] The maximum value among the first safety risk coefficient, the second safety risk coefficient, and the third safety risk coefficient of the action is taken as the safety risk coefficient of the corresponding action.

[0177] Based on the action initiator and action receiver of all actions, the total number of all action receivers of the monitored object is determined, and the ratio of the total number of all action receivers of the monitored object to the total number of all actions of the monitored object is taken as the fourth safety risk coefficient of the monitored object.

[0178] The maximum value among the safety risk coefficient of all actions of the monitored object and the fourth safety risk coefficient is taken as the current safety risk coefficient of the corresponding monitored object.

[0179] The beneficial effects of the above technology are that based on the multi-party action dynamic interaction model, the action initiator and action receiver of the action are determined, and combined with the limb action dynamic prediction result, the current safety risk coefficient of the monitored object with action interaction can be accurately analyzed.

[0180] Embodiment 8:

[0181] On the basis of embodiment 1, the safety risk early warning method for sports training, S4: based on the current safety risk coefficient, it is judged whether the safety early warning instruction needs to be sent, and the safety risk early warning result is obtained, including:

[0182] When the current safety risk coefficient exceeding the safety risk coefficient threshold exists, a safety warning instruction is issued.

[0183] When the current safety risk coefficient exceeding the safety risk coefficient threshold does not exist, the judgment result is reserved.

[0184] In this embodiment, the judgment result is the result obtained by judging whether the current safety risk coefficient exceeding the safety risk coefficient threshold exists.

[0185] In this embodiment, the safety risk coefficient threshold is a threshold of the current safety risk coefficient used for judging whether the safety warning instruction needs to be issued.

[0186] The above technology has the beneficial effect that by comparing the current safety risk coefficient with the safety risk coefficient threshold, the safety risk accidents possibly occurring in the sports training process can be timely warned.

[0187] Embodiment 9:

[0188] Based on the embodiment 1, the safety risk warning method for sports training further comprises:

[0189] S5: registering the source, the inspection record and the use record of the sports training equipment, and generating a safety account book of the sports training equipment;

[0190] S6: evaluating an equipment safety coefficient of the sports training equipment based on the safety account book of the sports training equipment, judging whether the safety warning instruction needs to be issued based on the equipment safety coefficient and an equipment safety coefficient threshold, and obtaining an equipment safety risk warning result.

[0191] In this embodiment, the source is the source manufacturer information and the purchase channel of the sports training equipment.

[0192] In this embodiment, the inspection record is a record containing the time and the person in charge of checking whether the sports training equipment has a fault.

[0193] In this embodiment, the use record is a record containing the user and the corresponding use time period of each sports training equipment.

[0194] In this embodiment, the safety account book is a management record containing a series of information that may affect the state of the sports training equipment, such as the source, the inspection record and the use record of the sports training equipment, for the purpose of realizing the safety management of the sports training equipment.

[0195] In this embodiment, the equipment safety coefficient of the sports training equipment is evaluated based on the safety account book of the sports training equipment, for example, including:

[0196] Determine the total use duration based on the use record in the inspection record in the security account book, and take the ratio of the total use duration and the preset use duration threshold as the equipment safety factor of the sports training equipment;

[0197] Or input the security account book into the security factor evaluation model trained in advance by using a large number of security account books with evaluated equipment safety factors, to obtain the equipment safety factor of the sports training equipment.

[0198] In this embodiment, the equipment safety factor is a numerical value representing the degree of safety risk of the sports training equipment currently existing, which is evaluated based on the security account book of the sports training equipment.

[0199] In this embodiment, the equipment safety factor threshold is a preset threshold of the equipment safety factor for determining whether to issue a safety warning instruction to prompt the user to pay attention to the state of the sports training equipment.

[0200] In this embodiment, determining whether to issue a safety warning instruction based on the equipment safety factor and the equipment safety factor threshold comprises:

[0201] Determining whether the equipment safety factor is not less than the equipment safety factor threshold, if yes, issuing the safety warning instruction, otherwise, retaining the corresponding determination result.

[0202] In this embodiment, the equipment safety risk warning result is the result obtained by determining whether to issue a safety warning instruction based on the equipment safety factor and the equipment safety factor threshold.

[0203] The beneficial effects of the above technology are: by generating the security account book of the sports training equipment, the intelligent safety management of the sports training equipment is realized, and based on the security account book, the equipment safety factor is analyzed, the equipment safety factor is compared with the equipment safety factor threshold to determine whether to issue a safety warning instruction, and then the safety accident in the sports training process caused by the state problem of the sports training equipment is pre-warned.

[0204] Embodiment 10:

[0205] The present application provides a safety risk warning system for sports training, referring to Figure 4 , comprising:

[0206] The action analysis module is configured to perform real-time body movement analysis on all monitoring objects based on the sports training monitoring video, and obtain a body movement analysis result.

[0207] The trajectory prediction module is configured to predict the body movement trajectory of the monitoring object based on the body movement analysis result, and obtain a body movement dynamic prediction result.

[0208] a risk assessment module, configured to evaluate a current safety risk coefficient of the monitored object based on the dynamic prediction result of the body movement;

[0209] a safety warning module, configured to judge whether a safety warning instruction needs to be sent based on the current safety risk coefficient, and obtain a safety risk warning result.

[0210] The above-mentioned technology has the beneficial effect that the body movement of the monitored object is analyzed based on the sports training monitored object, the body movement trajectory of the monitored object is predicted based on the analysis result of the body movement, the safety risk coefficient is evaluated based on the prediction result, and it is judged whether to perform safety warning. Without the need for human cost investment, the time lag and the defect of judgment error of the safety risk warning method of the traditional sports training are overcome, and timely, accurate and effective safety risk warning in the sports training process is realized.

[0211] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for early warning of safety risks in sports training, characterized in that, include: S1: Real-time limb movement analysis of all monitored objects is performed based on sports training monitoring videos to obtain limb movement analysis results; S2: Based on the results of limb movement analysis, predict the limb movement trajectory of the monitored object to obtain dynamic prediction results of limb movements; S3: Evaluate the current safety risk coefficient of the monitored object based on the dynamic prediction results of limb movements; S4: Determine whether a security warning command needs to be issued based on the current security risk coefficient, and obtain the security risk warning result; S1: Based on sports training monitoring videos, perform real-time limb movement analysis on all monitored objects to obtain limb movement analysis results, including: S101: Real-time acquisition of sports training monitoring video of the current sports training process; S102: Identify the difference region between the remaining video frames (excluding the first video frame) in the sports training monitoring video and their corresponding adjacent previous video frame. S103: Filter out multiple difference regions that are determined to belong to the same object from all difference regions in the sports training monitoring video, and treat the sequence of multiple consecutive difference regions determined to belong to the same object as the same object region sequence. S104: Calculate the similarity between each sequence of the same object region and each limb pose image in the limb image library, and take the sequence of the same object region with a similarity of not less than the similarity threshold as the limb action region sequence, and take all limb action region sequences as the limb action analysis result. Among them, multiple discrepancy regions identified as belonging to the same object were selected from all discrepancy regions in the sports training monitoring video, including: Based on the corner detection algorithm, all contour corners are detected in the contour of the difference region, the adjacent contour points of the contour corners are determined, and the tangent angle of the contour of the difference region at the adjacent contour points is determined. The length of the contour line between two adjacent contour points of the difference region is determined, and the ratio of the difference of the tangent angle at two adjacent contour points to the corresponding contour line length is used as the approximate curvature of the contour corner point. The line connecting the center point of the difference region and the adjacent contour points in the counterclockwise direction of the contour of the difference region is taken as the first line, and the line connecting the center point of the difference region and the adjacent contour points in the clockwise direction of the contour of the difference region is taken as the second line. The angle between the first and second lines corresponding to the contour corner points is taken as the approximate arc angle of the contour corner points. Based on the approximate curvature and approximate arc angle of all contour corner points in all discrepancy areas of sports training monitoring videos, multiple discrepancy areas that are determined to belong to the same object are filtered out from all discrepancy areas.

2. The safety risk early warning method for sports training according to claim 1, characterized in that, Based on the approximate curvature and approximate arc angle of all contour corner points in all discrepancy regions of the sports training monitoring video, multiple discrepancy regions that are determined to belong to the same object are selected from all discrepancy regions, including: The approximate arc curvature sequence is obtained by sorting the approximate arc angles of all contour corner points in the difference region in a counterclockwise direction, and the approximate arc angle sequence is obtained by sorting the approximate arc angles of all contour corner points in the difference region in a counterclockwise direction. Based on the average chromaticity, approximate arc curvature sequence, and approximate arc angle sequence of the difference region, the similarity between two difference regions belonging to two adjacent video frames is calculated. Two different regions with a similarity of not less than the similarity threshold are identified as belonging to the same object, thus obtaining a classifying result. Based on the results of the classification, multiple discrepancies that are determined to belong to the same object are selected from all discrepancies in the sports training monitoring video.

3. The safety risk early warning method for sports training according to claim 1, characterized in that, S2: Based on the limb movement analysis results, predict the limb movement trajectory of the monitored object to obtain dynamic prediction results of limb movements, including: S201: Perform contour recognition on video frames in sports training monitoring videos to obtain all contours in each video frame; S202: Based on all contours in the video frame and the contours of all differential regions in the limb movement region sequence, determine the monitoring object corresponding to each limb movement region sequence in the limb movement analysis results; S203: Based on the distribution location of all the different regions belonging to the same video frame in all limb movement region sequences of the same monitored object, determine the limb name of each limb movement region sequence; S204: Based on the coordinate representation of each differential region in the limb movement region sequence, determine the limb movement trajectory corresponding to each limb movement region sequence; S205: Based on the limb names and limb movement trajectories of all limb movement area sequences of the monitored object and the preset trajectory prediction model, predict the limb movement trajectory of the monitored object to obtain the dynamic prediction result of limb movement.

4. The safety risk early warning method for sports training according to claim 1, characterized in that, S3: Based on the dynamic prediction results of limb movements, assess the current safety risk coefficient of the monitored object, including: Determine whether there is any action interaction between all monitored objects in the sports training monitoring video. If so, build a multi-party action dynamic prediction interaction model for all monitored objects. Based on the multi-party action dynamic prediction interaction model, the current safety risk coefficient of the monitored object with action interaction is analyzed, and based on the limb action dynamic prediction results of the monitored object without action interaction, the current safety risk coefficient of the monitored object without action interaction is analyzed. Otherwise, the current safety risk coefficient of the monitored object is analyzed based on the dynamic prediction results of limb movement.

5. A safety risk early warning method for sports training according to claim 4, characterized in that, Based on the multi-party action dynamic prediction interaction model, the current security risk coefficient of the monitored object with action interaction is analyzed, including: The dynamic prediction and interaction model of multi-party actions determines the initiator and receiver of each action. Based on the dynamic prediction results of all limb movements and the initiators and recipients of all movements, the current safety risk coefficient of all monitored objects is analyzed.

6. The safety risk early warning method for sports training according to claim 1, characterized in that, S4: Based on the current security risk coefficient, determine whether a security warning command needs to be issued, and obtain the security risk warning result, including: When there is a current security risk coefficient that exceeds the security risk coefficient threshold, a security warning command will be issued. If there is no current security risk coefficient that exceeds the security risk coefficient threshold, the judgment result is retained.

7. The safety risk early warning method for sports training according to claim 1, characterized in that, Also includes: S5: Register the source, inspection records, and usage records of sports training equipment to generate a safety ledger for sports training equipment; S6: Based on the safety record of sports training equipment, assess the safety factor of the sports training equipment, determine whether a safety warning instruction needs to be issued based on the safety factor and the safety factor threshold, and obtain the equipment safety risk warning result.

8. A safety risk early warning system for sports training, characterized in that, A safety risk warning method for sports training, used to perform any one of claims 1 to 7, comprises: The motion analysis module is used to perform real-time limb motion analysis on all monitored objects based on sports training monitoring videos and obtain limb motion analysis results. The trajectory prediction module is used to predict the trajectory of the limb movements of the monitored object based on the limb movement analysis results, and obtain the dynamic prediction results of the limb movements. The risk assessment module is used to evaluate the current safety risk coefficient of the monitored object based on the dynamic prediction results of limb movements; The security early warning module is used to determine whether a security early warning command needs to be issued based on the current security risk coefficient, and to obtain the security risk early warning result.

Images